
Clustered Multimedia
NOD
:
Popularity-Based Article Prefetching and Placement
Y.J.Kim, T.U.Choi, K.O.Jung, Y.K.Kang, S.H.Park, Ki-Dong Chung
Department of Computer Science, Pusan National University, Korea
Abstract
According to the current profound development of
multimedia and networking technologies, the way people
communicate with, naturally, has evolved from a text-
oriented into a multimedia-oriented. But, the VOD system
technologies can't be easily applied to MNOD system
because of the difference of the basic data and its properties.
This is the reason why we propose the MNOD system.
NOD data composing news articles make a difference
to VOD data in terms of media type and size, life cycle of
articles and frequency of clients' interaction. Because of
NOD data's intrinsic characteristics, NOD article popularity
model may be different from that of VOD videos. Hence, we
analyze statistically the log data of one electronic newspaper
and show the popularity distribution of articles is different
form Zipf
s,
which is a popularity model of VOD data. We
propose a new article popularity model for NOD data, which
we call Multi-Selection Zipf distribution.
Also,
we propose
the article prefetching policy based on the popularity model
and life cycle model of NOD articles for increasing
performance in a
MNOD
system.
And, we consider the data placement on the MNOD
system with prefetch cache. The user requests can be
serviced from a cache
or
a disk. Consequently, in order
to
place a NOD data, we must consider the correlation between
disks and cache according to data popularity.
In
this paper,
we propose the data placement policy that considers both
disk-existence probability and cache-existence probability of
a data using data popularity
1.
Introduction
1.1
Motivation
As a VOD system gets more popularity these days, a
NOD(News On Demand) system supporting multimedia
services, we call a
MNOD
(Multimedia
NOD)
system in this
paper, will be popular as a news service in the near future.
Furthermore, the developments of computer network and
multimedia technology
are
making the multimedia-oriented
news services possible.
Previous works generally regarded VOD and MNOD
systems as an identical category from the viewpoint of using
multimedia data. Multimedia data usually need
high
bandwidth and massive storage space and have real-time
criteria. Therefore, we basically utilize the research results of
a VOD system. Clearly,
a
MNOD system can be categorized
into the same multimedia application as a VOD system, but
it has some intrinsic aspects that a VOD system doesn't
possess[4]. First, the articles of NOD service are made on
and off in a day while the VOD data is made once a half-
month or a month. Second, the more recent article is made,
the more users access the article. The good movies are
preferred to the others for a long time, but NOD data aren't.
The NOD articles, which have the best popularity, don't last
more than three days. Third, the number of articles that are
requested from each user is varied. Owing to
the
long length
of video data, users serviced in VOD system select just one
or two video data. But the length of NOD data is short, as a
result, users
can
select several data at a time. Finally, NOD
data
have the temporal and spatial access locality. The user
requests burst according to the access time and the kind of
articles.
So
the VOD system can't be fully applied to a
MNOD system because of these different characteristics.
In
short, we use the VOD system research results basically, but
will
revise and decorate it.
The previous work showed that the system with
prefetching has the better performance than the system
without prefetching[ 1, 151. But without considering the other
characterizations such as short-term life cycle, the difference
of user's access pattern according to time and etc, we can not
expect to
the
increased performance in the MNOD system.
The system with prefetching only based on data popularity
has the problem, which is data replacement happens too
often. Hence, using the analysis of electronic newspaper log-
files, we propose the popularity model and the life cycle
model of NOD articles. And we will suggest the article
prefetching policy based on time window according as the
models.
And then we consider the data placement on system
with prefetch cache. The user requests can be serviced from
a cache or a disk. Consequently, in order to place a MNOD
data, we must consider the correlation between disks and
cache according to data popularity.
In
this
paper, we will
propose
the
data placement policy that considers both disk-
existence probability and cache-existence probability of a
data using data popularity.
The remainder of
this
paper is organized as follow: In
section
2
the overall architecture of a proposed MNOD
system is presented briefly. Section
3
shows the popularity
model and life cycle model of NOD articles via the statistical
analysis of log files of electronic newspapers, and
in
section
4 we propose the article prefetching policy based on the
popularity and life cycle model of NOD articles. In section
5
194
1-9173/99
$10.00
0
1999
IEEE